You can check for yourself, not just take our word for it. We share the hope of creating a place we can trust and protecting privacy. Every page displays a clear notice at the top right.
Local processing, network access and storage are different questions. This page explains the limits of the network control and how to inspect behavior and clean up your content independently.
This page is not a privacy behavior audit of this site or a security certification. It does not promise zero network access, zero uploads of any information or absolute security across the site. It reflects our best efforts through technical architecture and code.
Network limits are your choice
Page connection limits block page API requests and form submissions without disabling local computation. They are off by default. They do not disconnect your device or block all browser network activity.
Blocks current-document menu refreshes, online authorization, time beacons and resource fetches. Local background computation is not network activity and is not disabled by this control. Features that need resource downloads may still be affected. Turn limits off and reload the affected page when needed. Requests are not silently allowed. Every page also displays a clear notice at the top right.
Switching saves your choice and reloads this information page. It applies only to the same origin in this browser. Other open pages do not change immediately: export your work before reloading them manually. Other domains, ports, browsers and downloaded offline files do not share the setting.
What is restricted, and what is not?
Restricts current-document connections governed by connect-src, including fetch, XHR, WebSocket, EventSource and sendBeacon, even to same-origin APIs, as well as form submissions. Local computation and data processing are not restricted.
Does not block navigation or scripts, styles, fonts and images loaded through resource elements such as script, link and img. Requests can still expose IP addresses, browser metadata or content in URLs. This is not a guarantee against all content uploads.
Local computation is not network activity, but background tasks may still make requests. This control does not currently guarantee coverage of those requests, unintegrated pages, independent embedded pages, other tabs, extensions, browser traffic or existing background services. It cannot protect against tampered website code or a compromised device.
The policy is chosen at page startup. Earlier requests cannot be recalled. Turning the preference off does not remove a policy already applied to a document: reload it. Storage failures are reported and must not be interpreted as limits being on.
The control does not delete or encrypt local data. The blocked-operation count covers only this page policy and may include local resource reads. It is neither an upload count nor a complete network log.
Separate the two paths
Content processing
Input → processing in your browser → preview or export
This describes local features, not every tool or operation. Read the current tool instructions and test its storage and network behavior.
Website network activity
Browser → page resources / menu / specific online services
Downloading code is different from uploading a document. Still, inspect the destination, URL, body and purpose of requests rather than relying on labels such as resource or API.
Document-tool feedback (Markdown editor, SparkPoint and HTML editor): only clicking Send feedback makes the browser post your rating, optional plain-text comment, page identifier, interface language and a random submission ID directly to /api/tool-feedback on the Pathadd origin. This helps the developer understand your experience. The server stores these fields and the receipt time. No login or email is required. Documents, screenshots, filenames and cookies are not attached, and no third-party feedback service is contacted. The random ID only deduplicates retries, not visitors across pages. Requests may appear in ordinary network logs. Connection limits prevent submission, not local editing or export. There is no automatic submission or background retry.
AI-to-Word feedback: only clicking Send feedback makes the browser post your rating, optional plain-text comment, tool identifier ai2word, interface language and a random submission ID directly to /api/tool-feedback on the Pathadd origin. The server stores these fields and the receipt time. Documents, screenshots, filenames and cookies are not attached, and no third-party service is contacted. The random ID only deduplicates retries. Connection limits prevent submission, not local editing or export. There is no automatic submission or background retry.
Public menu: pages request the tool directory. This is separate from document conversion.
Markdown-to-Word feedback: only when you click Send feedback does your browser post your rating, optional comment, tool identifier, interface language and a random submission ID to /api/tool-feedback on the Pathadd origin, so the developer can understand the export experience. No login or email is required. Documents, filenames and cookies are not attached, and no third-party feedback service is contacted. The server stores these fields and the receipt time. Comments are plain text; do not include sensitive information. The random ID only deduplicates retries of the same submission and does not track visitors across pages. Requests still pass through the site gateway and may appear in ordinary network logs. Connection limits prevent submission, not local export. There is no automatic submission or background retry.
Connection test: clicking Test connection sends a GET request to /gw/pathadd-check-test on this site origin, without user input, documents or browser cookies, and without following redirects. Connection limits should block this request. An HTTP response means this request reached the server, not that the API operation succeeded. The server can still log IP addresses, time and browser information, and restrict or block access by IP. The result applies only to this request, not all site traffic.
Screenshot recognition may need to download recognition software and language resources on first use. These downloads are not image uploads; check for any other requests.
OPRF recipient identities: during initial setup and text or file receipt, the browser sends a randomly blinded request to the same-origin Pathadd identity API. Receipt also sends the identity, delivery ID, delivery type and authenticator code. After server authorization and evaluation, the browser unblinds the result and applies Argon2id. Setup still submits the recipient public key and TOTP setup material. The server stores independently generated OPRF keys and TOTP seeds encrypted. Neither plaintext passphrases, file contents nor original filenames are sent, and no third-party OPRF service is contacted. Connection limits prevent setup and identity recovery. OPRF evaluation does not spend a claim; final claims are authorized separately.
Online authorization: some encrypted-delivery tools need identity, authorization, key-share and audit requests. Local encryption does not mean there is no network metadata.
Time locks (A Promise to the Future): decryption requires the browser to connect to a third-party public time-beacon service (drand) and obtain the matching beacon. Decryption then takes place locally.
Time-Sealed File Transfer: the browser first checks server eligibility with Pathadd. If access is not open, it stops without contacting drand. Otherwise, it fetches time-lock ciphertext from Pathadd without spending a claim, then connects directly to the public drand services (api.drand.sh / drand.cloudflare.com) to obtain and verify the scheduled beacon. The receiving identity is first restored through Pathadd OPRF and authenticator authorization without spending a claim; after verifying the public beacon and unlocking the time-locked share locally, the browser submits identity and an authenticator code to Pathadd again to claim the separate server share, spending one claim. Sending does not connect to drand. Beacon requests contain only the network identifier and round, not files, original filenames or keys. Decryption cannot complete when connections are restricted or the beacon is unavailable.
Browser environment recommendations
Use Chrome or Edge installed from an official source, and keep it up to date.
For sensitive content, prefer a separate browser profile without extensions. Some extensions can read or modify page content. Private browsing does not provide security isolation or encrypted storage.
Keep your browser and operating system up to date. Avoid handling sensitive content on public or untrusted computers.
Be aware of system clipboard history and cloud clipboard sync. Copying sensitive content may create additional copies or trigger syncing across devices.
Local processing still depends on the security of your browser, extensions and operating system. It can reduce content uploads, but cannot protect against a compromised browser or computer, and does not mean absolute security.
Check for yourself, rather than relying only on a promise
Method 1: disconnect the device yourself
Use non-sensitive test content first. Open the tool online and load required resources. Some features download resources only on first use.
Turn off Wi-Fi or networking in the operating system, or unplug the network cable. The website switch is not a substitute.
Run the operation, check the result and try exporting it. Success shows only that this loaded workflow can run offline, not that it never makes extra requests when online.
Clear sensitive input and close related tabs before reconnecting. Offline success alone cannot prove that nothing is queued for later transmission. Prefer a trusted, controlled offline environment for highly sensitive content.
Method 2: inspect the browser Network panel
In desktop Chrome or Edge, press F12 or Ctrl+Shift+I. Open Network, ensure recording is on, select All, and enable Preserve log if needed.
Observe page-loading requests, clear the list, and run an operation with fictional content. Do not filter only Fetch/XHR: inspect WS, Img, Doc, Other and other types.
Inspect destinations, full URLs, query parameters, payloads, headers and initiators. Check WebSocket messages too. A CSP-blocked entry does not mean a request was sent.
Turn page connection limits on, reload the tool and repeat the same test. Check which requests are blocked and which resource or background-task requests remain. Cache hits or observing no requests do not prove zero network use across the site.
Network is not device-level packet capture and may not cover browser, extension, other-page or all background activity. Do not publish screenshots or HAR files containing sensitive content, tokens, cookies or keys.
Clean up after handling sensitive content
Export and protect any backups you need first. Cleanup can permanently remove drafts, screenshot records, rules and restoration maps. Never share sensitive maps with substituted text.
Use the tool clear or local-data deletion controls first, and check draft-saving settings. Emptying an input field does not necessarily remove saved content.
Find this site data through the browser address-bar site information or site settings. Remove cookies, locally saved data, caches and background service data as needed. Menu names vary. Deleting site data also resets network limits and other preferences.
Ctrl+H usually opens history; it does not clean anything automatically. Ctrl+Shift+Delete usually opens the clear-browsing-data dialog. Check the time range and data types to avoid removing other sites data. Clearing history does not mean this site data is cleared.
Check downloads, exported files, the recycle bin and cloud-synced copies. Deleting a download-history entry does not delete the file. Normal deletion is not secure erasure.
Replace clipboard contents and check clipboard history and cross-device sync as needed. Close related tabs. Closing private windows usually clears session data, but does not automatically delete downloaded files or external copies.
Local processing does not mean nothing is stored. Local storage does not mean encrypted storage. These steps reduce retention; they do not guarantee zero traces, memory erasure or system-level security.